AI models improve COBOL translation, but system replacements remain risky
OpenAI says its latest AI models have improved at translating old software, including COBOL, into modern programming languages while preserving what the original programs do. The advance could help organisations update systems that support banking, government and other essential services, but experts caution that the results do not yet make replacing those systems safe.
COBOL has been in use since 1959; IBM estimates it underpins more than 40 per cent of online banking, 80 per cent of in-person credit card transactions and 95 per cent of ATM transactions. In OpenAI’s tests, GPT-6.1 Sol preserved behaviour in 87 per cent of cases at low reasoning effort and 98 per cent at higher effort, while other models scored much lower. Anthropic has also promoted AI-assisted modernisation, though it has not published comparable performance benchmarks.
In short
- AI tools show better results translating COBOL.
- OpenAI reports up to 98 per cent behaviour preservation.
- Experts say critical systems still need cautious modernisation.
AI Business Companies Government Markets Politics Technology
Read the full article at the source →
Originally published by The Verge as “Can AI help solve the COBOL code crisis?”.